Recruitment and data collection
This cross-sectional study included a simple random sample of 529 Italian-speaking adults affected by different chronic conditions. Patients were randomly selected and recruited through the online panel provided by Research Now (http://www.researchnow.com/en-US.aspx), a professional research institute with branches across the world. The panel covers a wide range of chronic conditions and counts more than 6.5 million registered subjects worldwide. Subjects belonging to the panel are carefully screened for authenticity and legitimacy via digital fingerprint and geo-IP-validation from the provider. Panel recruitment and data collection processes are compliant with national laws in each country where Research Now operates. All panelists are profiled on the basis of their socio-demographic, clinical and life-styles characteristics. The panel is certified to be statistically representative of all the covered populations. In our study, in order to guarantee data quality, respondents were asked to confirm their demographics (i.e. sex, date and place of birth, ethnicity, nationality, educational level, place of residency) and clinical condition (i.e. health status, chronic diagnosis, date of first diagnosis, prescribed medications) previously collected by the Research Now Panel. To be included in our study, patients had to be Italian, affected by one or more chronic conditions, aged over 18 years old, and of both genders. Patients with dementia, cognitive impairments, active psychiatric disorders, blindness, deafness, or insufficient Italian language skills to meaningfully answer to the questions or without informed consent were excluded from this study. All participants gave written informed consent before being enrolled in the study. Patients completed the PAM13-I questionnaire between October and December 2014. Ethic approval was obtained from the Ethics Committee of the Università Cattolica del Sacro Cuore, Milan (Italy).
Translation and cultural adaptation of the American PAM 13
After receiving permission from Insignia Health, Inc., PAM 13 - the American original version - [17] was translated as recommended by the World Health Organization’s procedures for cross-cultural validation and adaptation of self-report measures [36]. This method includes the following steps: forward translation, experts’ qualitative interviews, backward translation, pilot testing on patients (for checking the readability and understanding of items) and consensus about the final version (see Appendix 2). The forward and backward translation were performed by professionals who were familiar with the lexicon of the field, knowledgeable in both English and Italian cultures. A bilingual expert panel, composed by twelve individuals (experts in chronic care, health researchers, clinicians and translators), was convened to identify and resolve ambiguous expressions or concepts that could lead to misunderstanding. Discrepancies were discussed, consensus was achieved, and the cultural appropriateness of the translation was confirmed. Finally, a pilot testing of the scale was performed on fifteen chronic patients to investigate their understanding of the items and cognitive equivalence of the translation, followed by debriefing. PAM13-I was judged clear and acceptable and the final version was created by consensus.
Measures
PAM13-I
PAM13-I consists of 13 items on a Likert scale. According to the American version of PAM 13, each item has five response categories with scores from 1 to 5: (1) “Strongly Disagree”, (2) “Disagree”, (3) “Agree”, (4) “Strongly Agree” and (5) “Not Applicable”. The instrument design reflects the four stages of activation in a progressing difficulty of the items: level 1 (patients believe that their role is important: item 1 and 2), level 2 (patients have confidence and knowledge to take action: items 3–8), level 3 (taking action: items 9–11) and level 4 (staying on course under stress: items 12 and 13). According to Insigna Health Inc. guidelines, the raw scores were transformed through natural logarithm to achieve a better expression of the relative distance between the scores. Then, items were transformed to a standardized metric ranging from 0 to 100 (0 = lower activation; 100 = highest activation), to compare Italian results to the original data. The score was calculated by summing up the raw scores and mapping up the sum onto a scale of 0–100. A higher score of PAM13-I indicates a high level of patient activation.
Other measures
Age, gender, chronical disease (Asthma, Celiachia, Hypertension, Chronic Obstructive Pulmonary Disorder, Diabetes, Cardiovascular Disorder, Cancer, Chron, Fibromialgy, Coliteulcerosa, Lupus, Osteoatritis, Artritereumatoide, Hypercolesterolemia, Epatitis, Anemy, Allergy), marital status (divorced, married, single, widow, and widower), education (graduated, high school, middle school, and primary school), profession (employee, freelancer, student, retired, unemployed) and children (yes or no) were used as background variables.
Data analysis
According to COSMIN checklist [37] and to previous PAM 13 validation studies [24, 26], the statistical analysis was conducted in three main steps: missing data analysis, reliability analysis, and Rasch Model analysis.
The first phase regarded the data quality analysis. According to other validation studies [24, 25], participant who filled out less than 7 items on the PAM13-I questionnaire were excluded from validation study. Data were described for each item with frequencies and percentage of missing responses, response options (“Strongly Disagree”, “Disagree”, “Agree”, “Strongly Agree”, “Not Applicable”) and with several statistical indices (mean, median, standard deviation). An evaluation of floor and ceiling effects was also performed.
The second step regarded the reliability analysis. Internal consistency and reliability analysis were assessed using Cronbach’s α as well as item-rest correlation, inter-item correlation, average inter-item correlation. Cronbach’s α of 0.80 was defined as acceptable [38]. Item-rest correlation provided empirical evidence that each item was measuring the same construct measured by the other items included. A correlation value more than 0.3 indicates a moderate and valid correlation with the scale overall and, thus, the item should be not removed [39]. Average inter-item correlation is a subtype of internal consistency reliability, obtained by taking all of the items on a test that test the same construct, determining the correlation coefficient for each pair of items, and finally taking the average of all of these correlation coefficients. This final step yields the average inter-item correlation. An average inter-item correlation between 0.15-0.50 was considered acceptable [38].
In the third step, a Rasch Model (RM) was implemented to examine the PAM13-I psychometric properties. RM is useful to investigate unidimensionality of the construct (fundamental requisite of the summarization of the raw scores), the fit and the reliability of each item, and the differential item functioning.
The Rasch model assumes that the responses are affected by two different components that work independently [40]. The first component concerns the individual characteristics of the subjects and the other to the “displacement” of the generic item gathering the latent aspect of interest. The classical approach of RM [41] assumes that the response probability of each subject to a generic item depends on the level of the latent aspect (ability) and on the difficulty of the item. RM belongs to the family of IRT measurement models, which scale raw observed scores into linear reproducible measurements. Under the hypotheses that there are two different aspects (linked to subjects and items), acting in a separable manner, RM allows for constructing a single metric scale defining a ranking of Items and Person parameters. RM is designed to estimate the subject’s level on the latent trait, net of item characteristics, and items’ net of subjects. Let p
ijk
be the probability that unit i, with person parameter θ
i
, chooses the category k for evaluating the item j; it may be represented through a proper “link function” φ (θ
i
,β
j
) in the parameters θ
i
and β
j
, accounting, respectively, for personal and item characteristics. RM assumes the last relation to be of the logistic type. In the family of the polytomous models, we consider the Partial Credit Model (PCM) to our sample to examine model-data fit. PCM was chosen because the PAM13-I items had more than two response options and they showed different pattern of usage. Since it is reasonable to assume that the thresholds are not the same for all the items, i.e., each item has its own unique rating scale structure, the PCM appears the most appropriate model. The parameters of the model are estimated by the maximum likelihood method [41]. We performed PCA - part of PCA on Rasch residuals - in order to test the unidimensionality of the construct under investigation.
In the context of patient activation measure analysis, the parameters θ
i
and β
j
have a specific interpretation. The individual characteristic θ
i
, usually called “ability”, may be conceived as the individual activation: subjects with higher score in this subscale (Personal Location) will have a higher level of activation. The item characteristic β
j
, called “item difficulty” in the classical Rasch example, in this context represents the item propensity to obtain, by the respondents, systematically high or low scores when measuring the latent trait of interest. They reflect the level of relevance for a particular aspect measured by each item of the survey. In this way, it is possible to order the survey’s items basing on their different tendency in arousing the activation.
Unidimensionality of items composing PAM13-I was examined, using a Principal Component Analysis (PCA). The aim is to obtain one common factor that explains at least 30–40 % of the total variance. The other fundamental condition is that there are no other factors having eigenvalues greater than 3 [42]. Bartlett’s test of sphericity was performed to explore the factorability of the correlation matrix and Kaiser-Mayer-Olkin Index was calculated as a measure of sampling adequacy.
Local Independence of item was tested by a PCA conducted on the Rasch item measure residuals, in order to analyze the amount of unexplained variance and whether this unexplained variance indicates that there may be more than one dimension. Generally, Rasch-conforming data produces residual-factors with eigenvalues up to 2.0 [43]. Thus, if there is more than one contrast (factors) in the residuals, there may be a second dimension. Contrasts in the Rasch analysis of residuals contradict unidimensionality.
Two item fit mean square (MNSQ) statistics (infit and outfit) were computed to check whether the items fitted the expected model. MNSQ determines how well each item contributes to defining a single underlying construct. Infit is more sensitive to misfitting responses to items closest to the person's ability level, while outfit is more sensitive to misfitting items that are farther away. If the data fit the Rasch model, the fit statistics should be between 0.6 and 1.4. According [44], for clinical observations, the fit statistics could be between 0.5 and 1.5 [26].
The person separation index (PSI) is indicators of quality of measures. The PSI refers to the reproducibility of the measure location of the persons. A separation index, in its normalized form, of 0.80 or higher is considered reliable.
Differential Item functioning (DIF) was assessed in order to verify no relevant differences in the probability to endorse a certain item for subgroups, determined by gender, age and education. The scale should work irrespective of the group considered. Andersen’s Likelihood Ratio Test was performed for each item. To evaluate floor and ceiling effects we evaluated for each item the percentage incidence respectively of the lower level of the scale (Strongly Disagree) and of the upper one (Strongly Agree). Analyses were conducted with IBM SPSS 21.0 and R 3.0.3 (package eRm).